I built a self-hosted RAG system for Journalism — What Production Retrieval Taught Me

A builder shared lessons from Atlas, a self-hosted RAG system for journalism that powers cited answers, fact-checks, briefs, and repurposed content.

Atlas ingests live news, embeds content locally, and stores vectors in PostgreSQL with pgvector for a fully self-hosted pipeline.

In production, pure vector search missed news proper nouns, so the system switched to hybrid retrieval using full-text search plus rank fusion.

The team also found that batch embedding significantly improved ingestion speed and made the pipeline more practical to run reliably.